Closed InfluenceFunctional closed 1 year ago
Currently long epochs may mean we take many hours of training before ramping to maximum batch size.
Reimplement the batch sizer to increment after a certain number of steps rather than whole epochs.
same essential function as #68
Currently long epochs may mean we take many hours of training before ramping to maximum batch size.
Reimplement the batch sizer to increment after a certain number of steps rather than whole epochs.